falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
62275 Commits
32 Branches
792MB
Tree: 7b2b06eb43
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '7b2b06eb43'
${ noResults }
FFmpeg/libswscale/utils.c

2110 lines
77KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

  6.  * FFmpeg is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * FFmpeg is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with FFmpeg; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. #include "config.h"

  22. 

  23. #define _SVID_SOURCE // needed for MAP_ANONYMOUS

  24. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  25. #include <inttypes.h>

  26. #include <math.h>

  27. #include <stdio.h>

  28. #include <string.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

  31. #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. 

  40. #include "libavutil/attributes.h"

  41. #include "libavutil/avassert.h"

  42. #include "libavutil/avutil.h"

  43. #include "libavutil/bswap.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/intreadwrite.h"

  46. #include "libavutil/mathematics.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. #include "libavutil/ppc/cpu.h"

  50. #include "libavutil/x86/asm.h"

  51. #include "libavutil/x86/cpu.h"

  52. #include "rgb2rgb.h"

  53. #include "swscale.h"

  54. #include "swscale_internal.h"

  55. 

  56. static void handle_formats(SwsContext *c);

  57. 

  58. unsigned swscale_version(void)

  59. {

  60.     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);

  61.     return LIBSWSCALE_VERSION_INT;

  62. }

  63. 

  64. const char *swscale_configuration(void)

  65. {

  66.     return FFMPEG_CONFIGURATION;

  67. }

  68. 

  69. const char *swscale_license(void)

  70. {

  71. #define LICENSE_PREFIX "libswscale license: "

  72.     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;

  73. }

  74. 

  75. #define RET 0xC3 // near return opcode for x86

  76. 

  77. typedef struct FormatEntry {

  78.     uint8_t is_supported_in         :1;

  79.     uint8_t is_supported_out        :1;

  80.     uint8_t is_supported_endianness :1;

  81. } FormatEntry;

  82. 

  83. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

  84.     [AV_PIX_FMT_YUV420P]     = { 1, 1 },

  85.     [AV_PIX_FMT_YUYV422]     = { 1, 1 },

  86.     [AV_PIX_FMT_RGB24]       = { 1, 1 },

  87.     [AV_PIX_FMT_BGR24]       = { 1, 1 },

  88.     [AV_PIX_FMT_YUV422P]     = { 1, 1 },

  89.     [AV_PIX_FMT_YUV444P]     = { 1, 1 },

  90.     [AV_PIX_FMT_YUV410P]     = { 1, 1 },

  91.     [AV_PIX_FMT_YUV411P]     = { 1, 1 },

  92.     [AV_PIX_FMT_GRAY8]       = { 1, 1 },

  93.     [AV_PIX_FMT_MONOWHITE]   = { 1, 1 },

  94.     [AV_PIX_FMT_MONOBLACK]   = { 1, 1 },

  95.     [AV_PIX_FMT_PAL8]        = { 1, 0 },

  96.     [AV_PIX_FMT_YUVJ420P]    = { 1, 1 },

  97.     [AV_PIX_FMT_YUVJ411P]    = { 1, 1 },

  98.     [AV_PIX_FMT_YUVJ422P]    = { 1, 1 },

  99.     [AV_PIX_FMT_YUVJ444P]    = { 1, 1 },

  100.     [AV_PIX_FMT_UYVY422]     = { 1, 1 },

  101.     [AV_PIX_FMT_UYYVYY411]   = { 0, 0 },

  102.     [AV_PIX_FMT_BGR8]        = { 1, 1 },

  103.     [AV_PIX_FMT_BGR4]        = { 0, 1 },

  104.     [AV_PIX_FMT_BGR4_BYTE]   = { 1, 1 },

  105.     [AV_PIX_FMT_RGB8]        = { 1, 1 },

  106.     [AV_PIX_FMT_RGB4]        = { 0, 1 },

  107.     [AV_PIX_FMT_RGB4_BYTE]   = { 1, 1 },

  108.     [AV_PIX_FMT_NV12]        = { 1, 1 },

  109.     [AV_PIX_FMT_NV21]        = { 1, 1 },

  110.     [AV_PIX_FMT_ARGB]        = { 1, 1 },

  111.     [AV_PIX_FMT_RGBA]        = { 1, 1 },

  112.     [AV_PIX_FMT_ABGR]        = { 1, 1 },

  113.     [AV_PIX_FMT_BGRA]        = { 1, 1 },

  114.     [AV_PIX_FMT_0RGB]        = { 1, 1 },

  115.     [AV_PIX_FMT_RGB0]        = { 1, 1 },

  116.     [AV_PIX_FMT_0BGR]        = { 1, 1 },

  117.     [AV_PIX_FMT_BGR0]        = { 1, 1 },

  118.     [AV_PIX_FMT_GRAY16BE]    = { 1, 1 },

  119.     [AV_PIX_FMT_GRAY16LE]    = { 1, 1 },

  120.     [AV_PIX_FMT_YUV440P]     = { 1, 1 },

  121.     [AV_PIX_FMT_YUVJ440P]    = { 1, 1 },

  122.     [AV_PIX_FMT_YUVA420P]    = { 1, 1 },

  123.     [AV_PIX_FMT_YUVA422P]    = { 1, 1 },

  124.     [AV_PIX_FMT_YUVA444P]    = { 1, 1 },

  125.     [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },

  126.     [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },

  127.     [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },

  128.     [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },

  129.     [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },

  130.     [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },

  131.     [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },

  132.     [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },

  133.     [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },

  134.     [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },

  135.     [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },

  136.     [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },

  137.     [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },

  138.     [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },

  139.     [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },

  140.     [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },

  141.     [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },

  142.     [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },

  143.     [AV_PIX_FMT_RGB48BE]     = { 1, 1 },

  144.     [AV_PIX_FMT_RGB48LE]     = { 1, 1 },

  145.     [AV_PIX_FMT_RGBA64BE]    = { 1, 1, 1 },

  146.     [AV_PIX_FMT_RGBA64LE]    = { 1, 1, 1 },

  147.     [AV_PIX_FMT_RGB565BE]    = { 1, 1 },

  148.     [AV_PIX_FMT_RGB565LE]    = { 1, 1 },

  149.     [AV_PIX_FMT_RGB555BE]    = { 1, 1 },

  150.     [AV_PIX_FMT_RGB555LE]    = { 1, 1 },

  151.     [AV_PIX_FMT_BGR565BE]    = { 1, 1 },

  152.     [AV_PIX_FMT_BGR565LE]    = { 1, 1 },

  153.     [AV_PIX_FMT_BGR555BE]    = { 1, 1 },

  154.     [AV_PIX_FMT_BGR555LE]    = { 1, 1 },

  155.     [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },

  156.     [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },

  157.     [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },

  158.     [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },

  159.     [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },

  160.     [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },

  161.     [AV_PIX_FMT_RGB444LE]    = { 1, 1 },

  162.     [AV_PIX_FMT_RGB444BE]    = { 1, 1 },

  163.     [AV_PIX_FMT_BGR444LE]    = { 1, 1 },

  164.     [AV_PIX_FMT_BGR444BE]    = { 1, 1 },

  165.     [AV_PIX_FMT_Y400A]       = { 1, 0 },

  166.     [AV_PIX_FMT_BGR48BE]     = { 1, 1 },

  167.     [AV_PIX_FMT_BGR48LE]     = { 1, 1 },

  168.     [AV_PIX_FMT_BGRA64BE]    = { 1, 1, 1 },

  169.     [AV_PIX_FMT_BGRA64LE]    = { 1, 1, 1 },

  170.     [AV_PIX_FMT_YUV420P9BE]  = { 1, 1 },

  171.     [AV_PIX_FMT_YUV420P9LE]  = { 1, 1 },

  172.     [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },

  173.     [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },

  174.     [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },

  175.     [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },

  176.     [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },

  177.     [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },

  178.     [AV_PIX_FMT_YUV422P9BE]  = { 1, 1 },

  179.     [AV_PIX_FMT_YUV422P9LE]  = { 1, 1 },

  180.     [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },

  181.     [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },

  182.     [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },

  183.     [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },

  184.     [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },

  185.     [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },

  186.     [AV_PIX_FMT_YUV444P9BE]  = { 1, 1 },

  187.     [AV_PIX_FMT_YUV444P9LE]  = { 1, 1 },

  188.     [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },

  189.     [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },

  190.     [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },

  191.     [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },

  192.     [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },

  193.     [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },

  194.     [AV_PIX_FMT_GBRP]        = { 1, 1 },

  195.     [AV_PIX_FMT_GBRP9LE]     = { 1, 1 },

  196.     [AV_PIX_FMT_GBRP9BE]     = { 1, 1 },

  197.     [AV_PIX_FMT_GBRP10LE]    = { 1, 1 },

  198.     [AV_PIX_FMT_GBRP10BE]    = { 1, 1 },

  199.     [AV_PIX_FMT_GBRP12LE]    = { 1, 1 },

  200.     [AV_PIX_FMT_GBRP12BE]    = { 1, 1 },

  201.     [AV_PIX_FMT_GBRP14LE]    = { 1, 1 },

  202.     [AV_PIX_FMT_GBRP14BE]    = { 1, 1 },

  203.     [AV_PIX_FMT_GBRP16LE]    = { 1, 0 },

  204.     [AV_PIX_FMT_GBRP16BE]    = { 1, 0 },

  205.     [AV_PIX_FMT_XYZ12BE]     = { 1, 1, 1 },

  206.     [AV_PIX_FMT_XYZ12LE]     = { 1, 1, 1 },

  207.     [AV_PIX_FMT_GBRAP]       = { 1, 1 },

  208.     [AV_PIX_FMT_GBRAP16LE]   = { 1, 0 },

  209.     [AV_PIX_FMT_GBRAP16BE]   = { 1, 0 },

  210.     [AV_PIX_FMT_BAYER_BGGR8] = { 1, 0 },

  211.     [AV_PIX_FMT_BAYER_RGGB8] = { 1, 0 },

  212.     [AV_PIX_FMT_BAYER_GBRG8] = { 1, 0 },

  213.     [AV_PIX_FMT_BAYER_GRBG8] = { 1, 0 },

  214.     [AV_PIX_FMT_BAYER_BGGR16LE] = { 1, 0 },

  215.     [AV_PIX_FMT_BAYER_BGGR16BE] = { 1, 0 },

  216.     [AV_PIX_FMT_BAYER_RGGB16LE] = { 1, 0 },

  217.     [AV_PIX_FMT_BAYER_RGGB16BE] = { 1, 0 },

  218.     [AV_PIX_FMT_BAYER_GBRG16LE] = { 1, 0 },

  219.     [AV_PIX_FMT_BAYER_GBRG16BE] = { 1, 0 },

  220.     [AV_PIX_FMT_BAYER_GRBG16LE] = { 1, 0 },

  221.     [AV_PIX_FMT_BAYER_GRBG16BE] = { 1, 0 },

  222. };

  223. 

  224. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  225. {

  226.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  227.            format_entries[pix_fmt].is_supported_in : 0;

  228. }

  229. 

  230. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  231. {

  232.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  233.            format_entries[pix_fmt].is_supported_out : 0;

  234. }

  235. 

  236. int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)

  237. {

  238.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  239.            format_entries[pix_fmt].is_supported_endianness : 0;

  240. }

  241. 

  242. #if FF_API_SWS_FORMAT_NAME

  243. const char *sws_format_name(enum AVPixelFormat format)

  244. {

  245.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);

  246.     if (desc)

  247.         return desc->name;

  248.     else

  249.         return "Unknown format";

  250. }

  251. #endif

  252. 

  253. static double getSplineCoeff(double a, double b, double c, double d,

  254.                              double dist)

  255. {

  256.     if (dist <= 1.0)

  257.         return ((d * dist + c) * dist + b) * dist + a;

  258.     else

  259.         return getSplineCoeff(0.0,

  260.                                b + 2.0 * c + 3.0 * d,

  261.                                c + 3.0 * d,

  262.                               -b - 3.0 * c - 6.0 * d,

  263.                               dist - 1.0);

  264. }

  265. 

  266. static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)

  267. {

  268.     if (pos < 0) {

  269.         pos = (128 << chr_subsample) - 128;

  270.     }

  271.     pos += 128; // relative to ideal left edge

  272.     return pos >> chr_subsample;

  273. }

  274. 

  275. typedef struct {

  276.     int flag;                   ///< flag associated to the algorithm

  277.     const char *description;    ///< human-readable description

  278.     int size_factor;            ///< size factor used when initing the filters

  279. } ScaleAlgorithm;

  280. 

  281. static const ScaleAlgorithm scale_algorithms[] = {

  282.     { SWS_AREA,          "area averaging",                  1 /* downscale only, for upscale it is bilinear */ },

  283.     { SWS_BICUBIC,       "bicubic",                         4 },

  284.     { SWS_BICUBLIN,      "luma bicubic / chroma bilinear", -1 },

  285.     { SWS_BILINEAR,      "bilinear",                        2 },

  286.     { SWS_FAST_BILINEAR, "fast bilinear",                  -1 },

  287.     { SWS_GAUSS,         "Gaussian",                        8 /* infinite ;) */ },

  288.     { SWS_LANCZOS,       "Lanczos",                        -1 /* custom */ },

  289.     { SWS_POINT,         "nearest neighbor / point",       -1 },

  290.     { SWS_SINC,          "sinc",                           20 /* infinite ;) */ },

  291.     { SWS_SPLINE,        "bicubic spline",                 20 /* infinite :)*/ },

  292.     { SWS_X,             "experimental",                    8 },

  293. };

  294. 

  295. static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,

  296.                               int *outFilterSize, int xInc, int srcW,

  297.                               int dstW, int filterAlign, int one,

  298.                               int flags, int cpu_flags,

  299.                               SwsVector *srcFilter, SwsVector *dstFilter,

  300.                               double param[2], int srcPos, int dstPos)

  301. {

  302.     int i;

  303.     int filterSize;

  304.     int filter2Size;

  305.     int minFilterSize;

  306.     int64_t *filter    = NULL;

  307.     int64_t *filter2   = NULL;

  308.     const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));

  309.     int ret            = -1;

  310. 

  311.     emms_c(); // FIXME should not be required but IS (even for non-MMX versions)

  312. 

  313.     // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end

  314.     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);

  315. 

  316.     if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled

  317.         int i;

  318.         filterSize = 1;

  319.         FF_ALLOCZ_OR_GOTO(NULL, filter,

  320.                           dstW * sizeof(*filter) * filterSize, fail);

  321. 

  322.         for (i = 0; i < dstW; i++) {

  323.             filter[i * filterSize] = fone;

  324.             (*filterPos)[i]        = i;

  325.         }

  326.     } else if (flags & SWS_POINT) { // lame looking point sampling mode

  327.         int i;

  328.         int64_t xDstInSrc;

  329.         filterSize = 1;

  330.         FF_ALLOC_OR_GOTO(NULL, filter,

  331.                          dstW * sizeof(*filter) * filterSize, fail);

  332. 

  333.         xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);

  334.         for (i = 0; i < dstW; i++) {

  335.             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;

  336. 

  337.             (*filterPos)[i] = xx;

  338.             filter[i]       = fone;

  339.             xDstInSrc      += xInc;

  340.         }

  341.     } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||

  342.                (flags & SWS_FAST_BILINEAR)) { // bilinear upscale

  343.         int i;

  344.         int64_t xDstInSrc;

  345.         filterSize = 2;

  346.         FF_ALLOC_OR_GOTO(NULL, filter,

  347.                          dstW * sizeof(*filter) * filterSize, fail);

  348. 

  349.         xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);

  350.         for (i = 0; i < dstW; i++) {

  351.             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;

  352.             int j;

  353. 

  354.             (*filterPos)[i] = xx;

  355.             // bilinear upscale / linear interpolate / area averaging

  356.             for (j = 0; j < filterSize; j++) {

  357.                 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);

  358.                 if (coeff < 0)

  359.                     coeff = 0;

  360.                 filter[i * filterSize + j] = coeff;

  361.                 xx++;

  362.             }

  363.             xDstInSrc += xInc;

  364.         }

  365.     } else {

  366.         int64_t xDstInSrc;

  367.         int sizeFactor = -1;

  368. 

  369.         for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {

  370.             if (flags & scale_algorithms[i].flag && scale_algorithms[i].size_factor > 0) {

  371.                 sizeFactor = scale_algorithms[i].size_factor;

  372.                 break;

  373.             }

  374.         }

  375.         if (flags & SWS_LANCZOS)

  376.             sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;

  377.         av_assert0(sizeFactor > 0);

  378. 

  379.         if (xInc <= 1 << 16)

  380.             filterSize = 1 + sizeFactor;    // upscale

  381.         else

  382.             filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;

  383. 

  384.         filterSize = FFMIN(filterSize, srcW - 2);

  385.         filterSize = FFMAX(filterSize, 1);

  386. 

  387.         FF_ALLOC_OR_GOTO(NULL, filter,

  388.                          dstW * sizeof(*filter) * filterSize, fail);

  389. 

  390.         xDstInSrc = ((dstPos*(int64_t)xInc)>>7) - ((srcPos*0x10000LL)>>7);

  391.         for (i = 0; i < dstW; i++) {

  392.             int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);

  393.             int j;

  394.             (*filterPos)[i] = xx;

  395.             for (j = 0; j < filterSize; j++) {

  396.                 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;

  397.                 double floatd;

  398.                 int64_t coeff;

  399. 

  400.                 if (xInc > 1 << 16)

  401.                     d = d * dstW / srcW;

  402.                 floatd = d * (1.0 / (1 << 30));

  403. 

  404.                 if (flags & SWS_BICUBIC) {

  405.                     int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1 << 24);

  406.                     int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);

  407. 

  408.                     if (d >= 1LL << 31) {

  409.                         coeff = 0.0;

  410.                     } else {

  411.                         int64_t dd  = (d  * d) >> 30;

  412.                         int64_t ddd = (dd * d) >> 30;

  413. 

  414.                         if (d < 1LL << 30)

  415.                             coeff =  (12 * (1 << 24) -  9 * B - 6 * C) * ddd +

  416.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  417.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  418.                         else

  419.                             coeff =      (-B -  6 * C) * ddd +

  420.                                       (6 * B + 30 * C) * dd  +

  421.                                     (-12 * B - 48 * C) * d   +

  422.                                       (8 * B + 24 * C) * (1 << 30);

  423.                     }

  424.                     coeff /= (1LL<<54)/fone;

  425.                 }

  426. #if 0

  427.                 else if (flags & SWS_X) {

  428.                     double p  = param ? param * 0.01 : 0.3;

  429.                     coeff     = d ? sin(d * M_PI) / (d * M_PI) : 1.0;

  430.                     coeff    *= pow(2.0, -p * d * d);

  431.                 }

  432. #endif

  433.                 else if (flags & SWS_X) {

  434.                     double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;

  435.                     double c;

  436. 

  437.                     if (floatd < 1.0)

  438.                         c = cos(floatd * M_PI);

  439.                     else

  440.                         c = -1.0;

  441.                     if (c < 0.0)

  442.                         c = -pow(-c, A);

  443.                     else

  444.                         c = pow(c, A);

  445.                     coeff = (c * 0.5 + 0.5) * fone;

  446.                 } else if (flags & SWS_AREA) {

  447.                     int64_t d2 = d - (1 << 29);

  448.                     if (d2 * xInc < -(1LL << (29 + 16)))

  449.                         coeff = 1.0 * (1LL << (30 + 16));

  450.                     else if (d2 * xInc < (1LL << (29 + 16)))

  451.                         coeff = -d2 * xInc + (1LL << (29 + 16));

  452.                     else

  453.                         coeff = 0.0;

  454.                     coeff *= fone >> (30 + 16);

  455.                 } else if (flags & SWS_GAUSS) {

  456.                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;

  457.                     coeff = (pow(2.0, -p * floatd * floatd)) * fone;

  458.                 } else if (flags & SWS_SINC) {

  459.                     coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;

  460.                 } else if (flags & SWS_LANCZOS) {

  461.                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;

  462.                     coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /

  463.                              (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;

  464.                     if (floatd > p)

  465.                         coeff = 0;

  466.                 } else if (flags & SWS_BILINEAR) {

  467.                     coeff = (1 << 30) - d;

  468.                     if (coeff < 0)

  469.                         coeff = 0;

  470.                     coeff *= fone >> 30;

  471.                 } else if (flags & SWS_SPLINE) {

  472.                     double p = -2.196152422706632;

  473.                     coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;

  474.                 } else {

  475.                     av_assert0(0);

  476.                 }

  477. 

  478.                 filter[i * filterSize + j] = coeff;

  479.                 xx++;

  480.             }

  481.             xDstInSrc += 2 * xInc;

  482.         }

  483.     }

  484. 

  485.     /* apply src & dst Filter to filter -> filter2

  486.      * av_free(filter);

  487.      */

  488.     av_assert0(filterSize > 0);

  489.     filter2Size = filterSize;

  490.     if (srcFilter)

  491.         filter2Size += srcFilter->length - 1;

  492.     if (dstFilter)

  493.         filter2Size += dstFilter->length - 1;

  494.     av_assert0(filter2Size > 0);

  495.     FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);

  496. 

  497.     for (i = 0; i < dstW; i++) {

  498.         int j, k;

  499. 

  500.         if (srcFilter) {

  501.             for (k = 0; k < srcFilter->length; k++) {

  502.                 for (j = 0; j < filterSize; j++)

  503.                     filter2[i * filter2Size + k + j] +=

  504.                         srcFilter->coeff[k] * filter[i * filterSize + j];

  505.             }

  506.         } else {

  507.             for (j = 0; j < filterSize; j++)

  508.                 filter2[i * filter2Size + j] = filter[i * filterSize + j];

  509.         }

  510.         // FIXME dstFilter

  511. 

  512.         (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;

  513.     }

  514.     av_freep(&filter);

  515. 

  516.     /* try to reduce the filter-size (step1 find size and shift left) */

  517.     // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).

  518.     minFilterSize = 0;

  519.     for (i = dstW - 1; i >= 0; i--) {

  520.         int min = filter2Size;

  521.         int j;

  522.         int64_t cutOff = 0.0;

  523. 

  524.         /* get rid of near zero elements on the left by shifting left */

  525.         for (j = 0; j < filter2Size; j++) {

  526.             int k;

  527.             cutOff += FFABS(filter2[i * filter2Size]);

  528. 

  529.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  530.                 break;

  531. 

  532.             /* preserve monotonicity because the core can't handle the

  533.              * filter otherwise */

  534.             if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])

  535.                 break;

  536. 

  537.             // move filter coefficients left

  538.             for (k = 1; k < filter2Size; k++)

  539.                 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];

  540.             filter2[i * filter2Size + k - 1] = 0;

  541.             (*filterPos)[i]++;

  542.         }

  543. 

  544.         cutOff = 0;

  545.         /* count near zeros on the right */

  546.         for (j = filter2Size - 1; j > 0; j--) {

  547.             cutOff += FFABS(filter2[i * filter2Size + j]);

  548. 

  549.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  550.                 break;

  551.             min--;

  552.         }

  553. 

  554.         if (min > minFilterSize)

  555.             minFilterSize = min;

  556.     }

  557. 

  558.     if (PPC_ALTIVEC(cpu_flags)) {

  559.         // we can handle the special case 4, so we don't want to go the full 8

  560.         if (minFilterSize < 5)

  561.             filterAlign = 4;

  562. 

  563.         /* We really don't want to waste our time doing useless computation, so

  564.          * fall back on the scalar C code for very small filters.

  565.          * Vectorizing is worth it only if you have a decent-sized vector. */

  566.         if (minFilterSize < 3)

  567.             filterAlign = 1;

  568.     }

  569. 

  570.     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  571.         // special case for unscaled vertical filtering

  572.         if (minFilterSize == 1 && filterAlign == 2)

  573.             filterAlign = 1;

  574.     }

  575. 

  576.     av_assert0(minFilterSize > 0);

  577.     filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));

  578.     av_assert0(filterSize > 0);

  579.     filter = av_malloc(filterSize * dstW * sizeof(*filter));

  580.     if (!filter)

  581.         goto fail;

  582.     if (filterSize >= MAX_FILTER_SIZE * 16 /

  583.                       ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16)) {

  584.         av_log(NULL, AV_LOG_ERROR, "sws: filterSize %d is too large, try less extreme scaling or increase MAX_FILTER_SIZE and recompile\n", filterSize);

  585.         goto fail;

  586.     }

  587.     *outFilterSize = filterSize;

  588. 

  589.     if (flags & SWS_PRINT_INFO)

  590.         av_log(NULL, AV_LOG_VERBOSE,

  591.                "SwScaler: reducing / aligning filtersize %d -> %d\n",

  592.                filter2Size, filterSize);

  593.     /* try to reduce the filter-size (step2 reduce it) */

  594.     for (i = 0; i < dstW; i++) {

  595.         int j;

  596. 

  597.         for (j = 0; j < filterSize; j++) {

  598.             if (j >= filter2Size)

  599.                 filter[i * filterSize + j] = 0;

  600.             else

  601.                 filter[i * filterSize + j] = filter2[i * filter2Size + j];

  602.             if ((flags & SWS_BITEXACT) && j >= minFilterSize)

  603.                 filter[i * filterSize + j] = 0;

  604.         }

  605.     }

  606. 

  607.     // FIXME try to align filterPos if possible

  608. 

  609.     // fix borders

  610.     for (i = 0; i < dstW; i++) {

  611.         int j;

  612.         if ((*filterPos)[i] < 0) {

  613.             // move filter coefficients left to compensate for filterPos

  614.             for (j = 1; j < filterSize; j++) {

  615.                 int left = FFMAX(j + (*filterPos)[i], 0);

  616.                 filter[i * filterSize + left] += filter[i * filterSize + j];

  617.                 filter[i * filterSize + j]     = 0;

  618.             }

  619.             (*filterPos)[i]= 0;

  620.         }

  621. 

  622.         if ((*filterPos)[i] + filterSize > srcW) {

  623.             int shift = (*filterPos)[i] + filterSize - srcW;

  624.             // move filter coefficients right to compensate for filterPos

  625.             for (j = filterSize - 2; j >= 0; j--) {

  626.                 int right = FFMIN(j + shift, filterSize - 1);

  627.                 filter[i * filterSize + right] += filter[i * filterSize + j];

  628.                 filter[i * filterSize + j]      = 0;

  629.             }

  630.             (*filterPos)[i]= srcW - filterSize;

  631.         }

  632.     }

  633. 

  634.     // Note the +1 is for the MMX scaler which reads over the end

  635.     /* align at 16 for AltiVec (needed by hScale_altivec_real) */

  636.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

  637.                       *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);

  638. 

  639.     /* normalize & store in outFilter */

  640.     for (i = 0; i < dstW; i++) {

  641.         int j;

  642.         int64_t error = 0;

  643.         int64_t sum   = 0;

  644. 

  645.         for (j = 0; j < filterSize; j++) {

  646.             sum += filter[i * filterSize + j];

  647.         }

  648.         sum = (sum + one / 2) / one;

  649.         if (!sum) {

  650.             av_log(NULL, AV_LOG_WARNING, "SwScaler: zero vector in scaling\n");

  651.             sum = 1;

  652.         }

  653.         for (j = 0; j < *outFilterSize; j++) {

  654.             int64_t v = filter[i * filterSize + j] + error;

  655.             int intV  = ROUNDED_DIV(v, sum);

  656.             (*outFilter)[i * (*outFilterSize) + j] = intV;

  657.             error                                  = v - intV * sum;

  658.         }

  659.     }

  660. 

  661.     (*filterPos)[dstW + 0] =

  662.     (*filterPos)[dstW + 1] =

  663.     (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will

  664.                                                       * read over the end */

  665.     for (i = 0; i < *outFilterSize; i++) {

  666.         int k = (dstW - 1) * (*outFilterSize) + i;

  667.         (*outFilter)[k + 1 * (*outFilterSize)] =

  668.         (*outFilter)[k + 2 * (*outFilterSize)] =

  669.         (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];

  670.     }

  671. 

  672.     ret = 0;

  673. 

  674. fail:

  675.     if(ret < 0)

  676.         av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");

  677.     av_free(filter);

  678.     av_free(filter2);

  679.     return ret;

  680. }

  681. 

  682. #if HAVE_MMXEXT_INLINE

  683. static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,

  684.                                        int16_t *filter, int32_t *filterPos,

  685.                                        int numSplits)

  686. {

  687.     uint8_t *fragmentA;

  688.     x86_reg imm8OfPShufW1A;

  689.     x86_reg imm8OfPShufW2A;

  690.     x86_reg fragmentLengthA;

  691.     uint8_t *fragmentB;

  692.     x86_reg imm8OfPShufW1B;

  693.     x86_reg imm8OfPShufW2B;

  694.     x86_reg fragmentLengthB;

  695.     int fragmentPos;

  696. 

  697.     int xpos, i;

  698. 

  699.     // create an optimized horizontal scaling routine

  700.     /* This scaler is made of runtime-generated MMXEXT code using specially tuned

  701.      * pshufw instructions. For every four output pixels, if four input pixels

  702.      * are enough for the fast bilinear scaling, then a chunk of fragmentB is

  703.      * used. If five input pixels are needed, then a chunk of fragmentA is used.

  704.      */

  705. 

  706.     // code fragment

  707. 

  708.     __asm__ volatile (

  709.         "jmp                         9f                 \n\t"

  710.         // Begin

  711.         "0:                                             \n\t"

  712.         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"

  713.         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"

  714.         "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"

  715.         "punpcklbw                %%mm7, %%mm1          \n\t"

  716.         "punpcklbw                %%mm7, %%mm0          \n\t"

  717.         "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"

  718.         "1:                                             \n\t"

  719.         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"

  720.         "2:                                             \n\t"

  721.         "psubw                    %%mm1, %%mm0          \n\t"

  722.         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"

  723.         "pmullw                   %%mm3, %%mm0          \n\t"

  724.         "psllw                       $7, %%mm1          \n\t"

  725.         "paddw                    %%mm1, %%mm0          \n\t"

  726. 

  727.         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"

  728. 

  729.         "add                         $8, %%"REG_a"      \n\t"

  730.         // End

  731.         "9:                                             \n\t"

  732.         // "int $3                                         \n\t"

  733.         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"

  734.         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"

  735.         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"

  736.         "dec                         %1                 \n\t"

  737.         "dec                         %2                 \n\t"

  738.         "sub                         %0, %1             \n\t"

  739.         "sub                         %0, %2             \n\t"

  740.         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"

  741.         "sub                         %0, %3             \n\t"

  742. 

  743. 

  744.         : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),

  745.           "=r" (fragmentLengthA)

  746.         );

  747. 

  748.     __asm__ volatile (

  749.         "jmp                         9f                 \n\t"

  750.         // Begin

  751.         "0:                                             \n\t"

  752.         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"

  753.         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"

  754.         "punpcklbw                %%mm7, %%mm0          \n\t"

  755.         "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"

  756.         "1:                                             \n\t"

  757.         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"

  758.         "2:                                             \n\t"

  759.         "psubw                    %%mm1, %%mm0          \n\t"

  760.         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"

  761.         "pmullw                   %%mm3, %%mm0          \n\t"

  762.         "psllw                       $7, %%mm1          \n\t"

  763.         "paddw                    %%mm1, %%mm0          \n\t"

  764. 

  765.         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"

  766. 

  767.         "add                         $8, %%"REG_a"      \n\t"

  768.         // End

  769.         "9:                                             \n\t"

  770.         // "int                       $3                   \n\t"

  771.         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"

  772.         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"

  773.         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"

  774.         "dec                         %1                 \n\t"

  775.         "dec                         %2                 \n\t"

  776.         "sub                         %0, %1             \n\t"

  777.         "sub                         %0, %2             \n\t"

  778.         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"

  779.         "sub                         %0, %3             \n\t"

  780. 

  781. 

  782.         : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),

  783.           "=r" (fragmentLengthB)

  784.         );

  785. 

  786.     xpos        = 0; // lumXInc/2 - 0x8000; // difference between pixel centers

  787.     fragmentPos = 0;

  788. 

  789.     for (i = 0; i < dstW / numSplits; i++) {

  790.         int xx = xpos >> 16;

  791. 

  792.         if ((i & 3) == 0) {

  793.             int a                  = 0;

  794.             int b                  = ((xpos + xInc) >> 16) - xx;

  795.             int c                  = ((xpos + xInc * 2) >> 16) - xx;

  796.             int d                  = ((xpos + xInc * 3) >> 16) - xx;

  797.             int inc                = (d + 1 < 4);

  798.             uint8_t *fragment      = inc ? fragmentB : fragmentA;

  799.             x86_reg imm8OfPShufW1  = inc ? imm8OfPShufW1B : imm8OfPShufW1A;

  800.             x86_reg imm8OfPShufW2  = inc ? imm8OfPShufW2B : imm8OfPShufW2A;

  801.             x86_reg fragmentLength = inc ? fragmentLengthB : fragmentLengthA;

  802.             int maxShift           = 3 - (d + inc);

  803.             int shift              = 0;

  804. 

  805.             if (filterCode) {

  806.                 filter[i]        = ((xpos              & 0xFFFF) ^ 0xFFFF) >> 9;

  807.                 filter[i + 1]    = (((xpos + xInc)     & 0xFFFF) ^ 0xFFFF) >> 9;

  808.                 filter[i + 2]    = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;

  809.                 filter[i + 3]    = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;

  810.                 filterPos[i / 2] = xx;

  811. 

  812.                 memcpy(filterCode + fragmentPos, fragment, fragmentLength);

  813. 

  814.                 filterCode[fragmentPos + imm8OfPShufW1] =  (a + inc)       |

  815.                                                           ((b + inc) << 2) |

  816.                                                           ((c + inc) << 4) |

  817.                                                           ((d + inc) << 6);

  818.                 filterCode[fragmentPos + imm8OfPShufW2] =  a | (b << 2) |

  819.                                                                (c << 4) |

  820.                                                                (d << 6);

  821. 

  822.                 if (i + 4 - inc >= dstW)

  823.                     shift = maxShift;               // avoid overread

  824.                 else if ((filterPos[i / 2] & 3) <= maxShift)

  825.                     shift = filterPos[i / 2] & 3;   // align

  826. 

  827.                 if (shift && i >= shift) {

  828.                     filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;

  829.                     filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;

  830.                     filterPos[i / 2]                        -= shift;

  831.                 }

  832.             }

  833. 

  834.             fragmentPos += fragmentLength;

  835. 

  836.             if (filterCode)

  837.                 filterCode[fragmentPos] = RET;

  838.         }

  839.         xpos += xInc;

  840.     }

  841.     if (filterCode)

  842.         filterPos[((i / 2) + 1) & (~1)] = xpos >> 16;  // needed to jump to the next part

  843. 

  844.     return fragmentPos + 1;

  845. }

  846. #endif /* HAVE_MMXEXT_INLINE */

  847. 

  848. static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)

  849. {

  850.     int64_t W, V, Z, Cy, Cu, Cv;

  851.     int64_t vr =  table[0];

  852.     int64_t ub =  table[1];

  853.     int64_t ug = -table[2];

  854.     int64_t vg = -table[3];

  855.     int64_t ONE = 65536;

  856.     int64_t cy = ONE;

  857.     uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;

  858.     int i;

  859.     static const int8_t map[] = {

  860.     BY_IDX, GY_IDX, -1    , BY_IDX, BY_IDX, GY_IDX, -1    , BY_IDX,

  861.     RY_IDX, -1    , GY_IDX, RY_IDX, RY_IDX, -1    , GY_IDX, RY_IDX,

  862.     RY_IDX, GY_IDX, -1    , RY_IDX, RY_IDX, GY_IDX, -1    , RY_IDX,

  863.     BY_IDX, -1    , GY_IDX, BY_IDX, BY_IDX, -1    , GY_IDX, BY_IDX,

  864.     BU_IDX, GU_IDX, -1    , BU_IDX, BU_IDX, GU_IDX, -1    , BU_IDX,

  865.     RU_IDX, -1    , GU_IDX, RU_IDX, RU_IDX, -1    , GU_IDX, RU_IDX,

  866.     RU_IDX, GU_IDX, -1    , RU_IDX, RU_IDX, GU_IDX, -1    , RU_IDX,

  867.     BU_IDX, -1    , GU_IDX, BU_IDX, BU_IDX, -1    , GU_IDX, BU_IDX,

  868.     BV_IDX, GV_IDX, -1    , BV_IDX, BV_IDX, GV_IDX, -1    , BV_IDX,

  869.     RV_IDX, -1    , GV_IDX, RV_IDX, RV_IDX, -1    , GV_IDX, RV_IDX,

  870.     RV_IDX, GV_IDX, -1    , RV_IDX, RV_IDX, GV_IDX, -1    , RV_IDX,

  871.     BV_IDX, -1    , GV_IDX, BV_IDX, BV_IDX, -1    , GV_IDX, BV_IDX,

  872.     RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,

  873.     BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,

  874.     GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    ,

  875.     -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX,

  876.     RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,

  877.     BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,

  878.     GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    ,

  879.     -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX,

  880.     RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,

  881.     BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,

  882.     GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    ,

  883.     -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, //23

  884.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //24

  885.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //25

  886.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //26

  887.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //27

  888.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //28

  889.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //29

  890.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //30

  891.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //31

  892.     BY_IDX, GY_IDX, RY_IDX, -1    , -1    , -1    , -1    , -1    , //32

  893.     BU_IDX, GU_IDX, RU_IDX, -1    , -1    , -1    , -1    , -1    , //33

  894.     BV_IDX, GV_IDX, RV_IDX, -1    , -1    , -1    , -1    , -1    , //34

  895.     };

  896. 

  897.     dstRange = 0; //FIXME range = 1 is handled elsewhere

  898. 

  899.     if (!dstRange) {

  900.         cy = cy * 255 / 219;

  901.     } else {

  902.         vr = vr * 224 / 255;

  903.         ub = ub * 224 / 255;

  904.         ug = ug * 224 / 255;

  905.         vg = vg * 224 / 255;

  906.     }

  907.     W = ROUNDED_DIV(ONE*ONE*ug, ub);

  908.     V = ROUNDED_DIV(ONE*ONE*vg, vr);

  909.     Z = ONE*ONE-W-V;

  910. 

  911.     Cy = ROUNDED_DIV(cy*Z, ONE);

  912.     Cu = ROUNDED_DIV(ub*Z, ONE);

  913.     Cv = ROUNDED_DIV(vr*Z, ONE);

  914. 

  915.     c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V        , Cy);

  916.     c->input_rgb2yuv_table[GY_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cy);

  917.     c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W        , Cy);

  918. 

  919.     c->input_rgb2yuv_table[RU_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V        , Cu);

  920.     c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cu);

  921.     c->input_rgb2yuv_table[BU_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W)    , Cu);

  922. 

  923.     c->input_rgb2yuv_table[RV_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z)    , Cv);

  924.     c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cv);

  925.     c->input_rgb2yuv_table[BV_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W        , Cv);

  926. 

  927.     if(/*!dstRange && */!memcmp(table, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], sizeof(ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]))) {

  928.         c->input_rgb2yuv_table[BY_IDX] =  ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  929.         c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  930.         c->input_rgb2yuv_table[BU_IDX] =  ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  931.         c->input_rgb2yuv_table[GY_IDX] =  ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  932.         c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  933.         c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  934.         c->input_rgb2yuv_table[RY_IDX] =  ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  935.         c->input_rgb2yuv_table[RV_IDX] =  ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  936.         c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  937.     }

  938.     for(i=0; i<FF_ARRAY_ELEMS(map); i++)

  939.         AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);

  940. }

  941. 

  942. static void fill_xyztables(struct SwsContext *c)

  943. {

  944.     int i;

  945.     double xyzgamma = XYZ_GAMMA;

  946.     double rgbgamma = 1.0 / RGB_GAMMA;

  947.     double xyzgammainv = 1.0 / XYZ_GAMMA;

  948.     double rgbgammainv = RGB_GAMMA;

  949.     static const int16_t xyz2rgb_matrix[3][4] = {

  950.         {13270, -6295, -2041},

  951.         {-3969,  7682,   170},

  952.         {  228,  -835,  4329} };

  953.     static const int16_t rgb2xyz_matrix[3][4] = {

  954.         {1689, 1464,  739},

  955.         { 871, 2929,  296},

  956.         {  79,  488, 3891} };

  957.     static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096], xyzgammainv_tab[4096], rgbgammainv_tab[4096];

  958. 

  959.     memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));

  960.     memcpy(c->rgb2xyz_matrix, rgb2xyz_matrix, sizeof(c->rgb2xyz_matrix));

  961.     c->xyzgamma = xyzgamma_tab;

  962.     c->rgbgamma = rgbgamma_tab;

  963.     c->xyzgammainv = xyzgammainv_tab;

  964.     c->rgbgammainv = rgbgammainv_tab;

  965. 

  966.     if (rgbgamma_tab[4095])

  967.         return;

  968. 

  969.     /* set gamma vectors */

  970.     for (i = 0; i < 4096; i++) {

  971.         xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);

  972.         rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);

  973.         xyzgammainv_tab[i] = lrint(pow(i / 4095.0, xyzgammainv) * 4095.0);

  974.         rgbgammainv_tab[i] = lrint(pow(i / 4095.0, rgbgammainv) * 4095.0);

  975.     }

  976. }

  977. 

  978. int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],

  979.                              int srcRange, const int table[4], int dstRange,

  980.                              int brightness, int contrast, int saturation)

  981. {

  982.     const AVPixFmtDescriptor *desc_dst;

  983.     const AVPixFmtDescriptor *desc_src;

  984.     memmove(c->srcColorspaceTable, inv_table, sizeof(int) * 4);

  985.     memmove(c->dstColorspaceTable, table, sizeof(int) * 4);

  986. 

  987.     handle_formats(c);

  988.     desc_dst = av_pix_fmt_desc_get(c->dstFormat);

  989.     desc_src = av_pix_fmt_desc_get(c->srcFormat);

  990. 

  991.     if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))

  992.         dstRange = 0;

  993.     if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))

  994.         srcRange = 0;

  995. 

  996.     c->brightness = brightness;

  997.     c->contrast   = contrast;

  998.     c->saturation = saturation;

  999.     c->srcRange   = srcRange;

  1000.     c->dstRange   = dstRange;

  1001. 

  1002.     if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))

  1003.         return -1;

  1004. 

  1005.     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);

  1006.     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);

  1007. 

  1008.     if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {

  1009.         ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,

  1010.                                  contrast, saturation);

  1011.         // FIXME factorize

  1012. 

  1013.         if (ARCH_PPC)

  1014.             ff_yuv2rgb_init_tables_ppc(c, inv_table, brightness,

  1015.                                        contrast, saturation);

  1016.     }

  1017. 

  1018.     fill_rgb2yuv_table(c, table, dstRange);

  1019. 

  1020.     return 0;

  1021. }

  1022. 

  1023. int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,

  1024.                              int *srcRange, int **table, int *dstRange,

  1025.                              int *brightness, int *contrast, int *saturation)

  1026. {

  1027.     if (!c )

  1028.         return -1;

  1029. 

  1030.     *inv_table  = c->srcColorspaceTable;

  1031.     *table      = c->dstColorspaceTable;

  1032.     *srcRange   = c->srcRange;

  1033.     *dstRange   = c->dstRange;

  1034.     *brightness = c->brightness;

  1035.     *contrast   = c->contrast;

  1036.     *saturation = c->saturation;

  1037. 

  1038.     return 0;

  1039. }

  1040. 

  1041. static int handle_jpeg(enum AVPixelFormat *format)

  1042. {

  1043.     switch (*format) {

  1044.     case AV_PIX_FMT_YUVJ420P:

  1045.         *format = AV_PIX_FMT_YUV420P;

  1046.         return 1;

  1047.     case AV_PIX_FMT_YUVJ411P:

  1048.         *format = AV_PIX_FMT_YUV411P;

  1049.         return 1;

  1050.     case AV_PIX_FMT_YUVJ422P:

  1051.         *format = AV_PIX_FMT_YUV422P;

  1052.         return 1;

  1053.     case AV_PIX_FMT_YUVJ444P:

  1054.         *format = AV_PIX_FMT_YUV444P;

  1055.         return 1;

  1056.     case AV_PIX_FMT_YUVJ440P:

  1057.         *format = AV_PIX_FMT_YUV440P;

  1058.         return 1;

  1059.     case AV_PIX_FMT_GRAY8:

  1060.     case AV_PIX_FMT_GRAY16LE:

  1061.     case AV_PIX_FMT_GRAY16BE:

  1062.         return 1;

  1063.     default:

  1064.         return 0;

  1065.     }

  1066. }

  1067. 

  1068. static int handle_0alpha(enum AVPixelFormat *format)

  1069. {

  1070.     switch (*format) {

  1071.     case AV_PIX_FMT_0BGR    : *format = AV_PIX_FMT_ABGR   ; return 1;

  1072.     case AV_PIX_FMT_BGR0    : *format = AV_PIX_FMT_BGRA   ; return 4;

  1073.     case AV_PIX_FMT_0RGB    : *format = AV_PIX_FMT_ARGB   ; return 1;

  1074.     case AV_PIX_FMT_RGB0    : *format = AV_PIX_FMT_RGBA   ; return 4;

  1075.     default:                                          return 0;

  1076.     }

  1077. }

  1078. 

  1079. static int handle_xyz(enum AVPixelFormat *format)

  1080. {

  1081.     switch (*format) {

  1082.     case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;

  1083.     case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;

  1084.     default:                                                return 0;

  1085.     }

  1086. }

  1087. 

  1088. static void handle_formats(SwsContext *c)

  1089. {

  1090.     c->src0Alpha |= handle_0alpha(&c->srcFormat);

  1091.     c->dst0Alpha |= handle_0alpha(&c->dstFormat);

  1092.     c->srcXYZ    |= handle_xyz(&c->srcFormat);

  1093.     c->dstXYZ    |= handle_xyz(&c->dstFormat);

  1094.     if (c->srcXYZ || c->dstXYZ)

  1095.         fill_xyztables(c);

  1096. }

  1097. 

  1098. SwsContext *sws_alloc_context(void)

  1099. {

  1100.     SwsContext *c = av_mallocz(sizeof(SwsContext));

  1101. 

  1102.     if (c) {

  1103.         c->av_class = &sws_context_class;

  1104.         av_opt_set_defaults(c);

  1105.     }

  1106. 

  1107.     return c;

  1108. }

  1109. 

  1110. av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,

  1111.                              SwsFilter *dstFilter)

  1112. {

  1113.     int i, j;

  1114.     int usesVFilter, usesHFilter;

  1115.     int unscaled;

  1116.     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };

  1117.     int srcW              = c->srcW;

  1118.     int srcH              = c->srcH;

  1119.     int dstW              = c->dstW;

  1120.     int dstH              = c->dstH;

  1121.     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);

  1122.     int flags, cpu_flags;

  1123.     enum AVPixelFormat srcFormat = c->srcFormat;

  1124.     enum AVPixelFormat dstFormat = c->dstFormat;

  1125.     const AVPixFmtDescriptor *desc_src;

  1126.     const AVPixFmtDescriptor *desc_dst;

  1127. 

  1128.     cpu_flags = av_get_cpu_flags();

  1129.     flags     = c->flags;

  1130.     emms_c();

  1131.     if (!rgb15to16)

  1132.         sws_rgb2rgb_init();

  1133. 

  1134.     unscaled = (srcW == dstW && srcH == dstH);

  1135. 

  1136.     c->srcRange |= handle_jpeg(&c->srcFormat);

  1137.     c->dstRange |= handle_jpeg(&c->dstFormat);

  1138. 

  1139.     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)

  1140.         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");

  1141. 

  1142.     if (!c->contrast && !c->saturation && !c->dstFormatBpp)

  1143.         sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1144.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1145.                                  c->dstRange, 0, 1 << 16, 1 << 16);

  1146. 

  1147.     handle_formats(c);

  1148.     srcFormat = c->srcFormat;

  1149.     dstFormat = c->dstFormat;

  1150.     desc_src = av_pix_fmt_desc_get(srcFormat);

  1151.     desc_dst = av_pix_fmt_desc_get(dstFormat);

  1152. 

  1153.     if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&

  1154.           av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {

  1155.     if (!sws_isSupportedInput(srcFormat)) {

  1156.         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",

  1157.                av_get_pix_fmt_name(srcFormat));

  1158.         return AVERROR(EINVAL);

  1159.     }

  1160.     if (!sws_isSupportedOutput(dstFormat)) {

  1161.         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",

  1162.                av_get_pix_fmt_name(dstFormat));

  1163.         return AVERROR(EINVAL);

  1164.     }

  1165.     }

  1166. 

  1167.     i = flags & (SWS_POINT         |

  1168.                  SWS_AREA          |

  1169.                  SWS_BILINEAR      |

  1170.                  SWS_FAST_BILINEAR |

  1171.                  SWS_BICUBIC       |

  1172.                  SWS_X             |

  1173.                  SWS_GAUSS         |

  1174.                  SWS_LANCZOS       |

  1175.                  SWS_SINC          |

  1176.                  SWS_SPLINE        |

  1177.                  SWS_BICUBLIN);

  1178. 

  1179.     /* provide a default scaler if not set by caller */

  1180.     if (!i) {

  1181.         if (dstW < srcW && dstH < srcH)

  1182.             flags |= SWS_BICUBIC;

  1183.         else if (dstW > srcW && dstH > srcH)

  1184.             flags |= SWS_BICUBIC;

  1185.         else

  1186.             flags |= SWS_BICUBIC;

  1187.         c->flags = flags;

  1188.     } else if (i & (i - 1)) {

  1189.         av_log(c, AV_LOG_ERROR,

  1190.                "Exactly one scaler algorithm must be chosen, got %X\n", i);

  1191.         return AVERROR(EINVAL);

  1192.     }

  1193.     /* sanity check */

  1194.     if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {

  1195.         /* FIXME check if these are enough and try to lower them after

  1196.          * fixing the relevant parts of the code */

  1197.         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",

  1198.                srcW, srcH, dstW, dstH);

  1199.         return AVERROR(EINVAL);

  1200.     }

  1201. 

  1202.     if (!dstFilter)

  1203.         dstFilter = &dummyFilter;

  1204.     if (!srcFilter)

  1205.         srcFilter = &dummyFilter;

  1206. 

  1207.     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;

  1208.     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;

  1209.     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);

  1210.     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);

  1211.     c->vRounder     = 4 * 0x0001000100010001ULL;

  1212. 

  1213.     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||

  1214.                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||

  1215.                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||

  1216.                   (dstFilter->chrV && dstFilter->chrV->length > 1);

  1217.     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||

  1218.                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||

  1219.                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||

  1220.                   (dstFilter->chrH && dstFilter->chrH->length > 1);

  1221. 

  1222.     av_pix_fmt_get_chroma_sub_sample(srcFormat, &c->chrSrcHSubSample, &c->chrSrcVSubSample);

  1223.     av_pix_fmt_get_chroma_sub_sample(dstFormat, &c->chrDstHSubSample, &c->chrDstVSubSample);

  1224. 

  1225.     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {

  1226.         if (dstW&1) {

  1227.             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");

  1228.             flags |= SWS_FULL_CHR_H_INT;

  1229.             c->flags = flags;

  1230.         }

  1231. 

  1232.         if (   c->chrSrcHSubSample == 0

  1233.             && c->chrSrcVSubSample == 0

  1234.             && c->dither != SWS_DITHER_BAYER //SWS_FULL_CHR_H_INT is currently not supported with SWS_DITHER_BAYER

  1235.             && !(c->flags & SWS_FAST_BILINEAR)

  1236.         ) {

  1237.             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to input having non subsampled chroma\n");

  1238.             flags |= SWS_FULL_CHR_H_INT;

  1239.             c->flags = flags;

  1240.         }

  1241.     }

  1242. 

  1243.     if (c->dither == SWS_DITHER_AUTO) {

  1244.         if (flags & SWS_ERROR_DIFFUSION)

  1245.             c->dither = SWS_DITHER_ED;

  1246.     }

  1247. 

  1248.     if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||

  1249.        dstFormat == AV_PIX_FMT_RGB4_BYTE ||

  1250.        dstFormat == AV_PIX_FMT_BGR8 ||

  1251.        dstFormat == AV_PIX_FMT_RGB8) {

  1252.         if (c->dither == SWS_DITHER_AUTO)

  1253.             c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;

  1254.         if (!(flags & SWS_FULL_CHR_H_INT)) {

  1255.             if (c->dither == SWS_DITHER_ED || c->dither == SWS_DITHER_A_DITHER || c->dither == SWS_DITHER_X_DITHER) {

  1256.                 av_log(c, AV_LOG_DEBUG,

  1257.                     "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",

  1258.                     av_get_pix_fmt_name(dstFormat));

  1259.                 flags   |= SWS_FULL_CHR_H_INT;

  1260.                 c->flags = flags;

  1261.             }

  1262.         }

  1263.         if (flags & SWS_FULL_CHR_H_INT) {

  1264.             if (c->dither == SWS_DITHER_BAYER) {

  1265.                 av_log(c, AV_LOG_DEBUG,

  1266.                     "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",

  1267.                     av_get_pix_fmt_name(dstFormat));

  1268.                 c->dither = SWS_DITHER_ED;

  1269.             }

  1270.         }

  1271.     }

  1272.     if (isPlanarRGB(dstFormat)) {

  1273.         if (!(flags & SWS_FULL_CHR_H_INT)) {

  1274.             av_log(c, AV_LOG_DEBUG,

  1275.                    "%s output is not supported with half chroma resolution, switching to full\n",

  1276.                    av_get_pix_fmt_name(dstFormat));

  1277.             flags   |= SWS_FULL_CHR_H_INT;

  1278.             c->flags = flags;

  1279.         }

  1280.     }

  1281. 

  1282.     /* reuse chroma for 2 pixels RGB/BGR unless user wants full

  1283.      * chroma interpolation */

  1284.     if (flags & SWS_FULL_CHR_H_INT &&

  1285.         isAnyRGB(dstFormat)        &&

  1286.         !isPlanarRGB(dstFormat)    &&

  1287.         dstFormat != AV_PIX_FMT_RGBA  &&

  1288.         dstFormat != AV_PIX_FMT_ARGB  &&

  1289.         dstFormat != AV_PIX_FMT_BGRA  &&

  1290.         dstFormat != AV_PIX_FMT_ABGR  &&

  1291.         dstFormat != AV_PIX_FMT_RGB24 &&

  1292.         dstFormat != AV_PIX_FMT_BGR24 &&

  1293.         dstFormat != AV_PIX_FMT_BGR4_BYTE &&

  1294.         dstFormat != AV_PIX_FMT_RGB4_BYTE &&

  1295.         dstFormat != AV_PIX_FMT_BGR8 &&

  1296.         dstFormat != AV_PIX_FMT_RGB8

  1297.     ) {

  1298.         av_log(c, AV_LOG_WARNING,

  1299.                "full chroma interpolation for destination format '%s' not yet implemented\n",

  1300.                av_get_pix_fmt_name(dstFormat));

  1301.         flags   &= ~SWS_FULL_CHR_H_INT;

  1302.         c->flags = flags;

  1303.     }

  1304.     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))

  1305.         c->chrDstHSubSample = 1;

  1306. 

  1307.     // drop some chroma lines if the user wants it

  1308.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  1309.                            SWS_SRC_V_CHR_DROP_SHIFT;

  1310.     c->chrSrcVSubSample += c->vChrDrop;

  1311. 

  1312.     /* drop every other pixel for chroma calculation unless user

  1313.      * wants full chroma */

  1314.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  1315.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  1316.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  1317.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  1318.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  1319.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  1320.         srcFormat != AV_PIX_FMT_GBRP12BE  && srcFormat != AV_PIX_FMT_GBRP12LE &&

  1321.         srcFormat != AV_PIX_FMT_GBRP14BE  && srcFormat != AV_PIX_FMT_GBRP14LE &&

  1322.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

  1323.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  1324.          (flags & SWS_FAST_BILINEAR)))

  1325.         c->chrSrcHSubSample = 1;

  1326. 

  1327.     // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.

  1328.     c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);

  1329.     c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);

  1330.     c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);

  1331.     c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);

  1332. 

  1333.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1334. 

  1335.     /* unscaled special cases */

  1336.     if (unscaled && !usesHFilter && !usesVFilter &&

  1337.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1338.         ff_get_unscaled_swscale(c);

  1339. 

  1340.         if (c->swscale) {

  1341.             if (flags & SWS_PRINT_INFO)

  1342.                 av_log(c, AV_LOG_INFO,

  1343.                        "using unscaled %s -> %s special converter\n",

  1344.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  1345.             return 0;

  1346.         }

  1347.     }

  1348. 

  1349.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1350.     if (c->srcBpc < 8)

  1351.         c->srcBpc = 8;

  1352.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1353.     if (c->dstBpc < 8)

  1354.         c->dstBpc = 8;

  1355.     if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)

  1356.         c->srcBpc = 16;

  1357.     if (c->dstBpc == 16)

  1358.         dst_stride <<= 1;

  1359. 

  1360.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {

  1361.         c->canMMXEXTBeUsed = dstW >= srcW && (dstW & 31) == 0 &&

  1362.                              c->chrDstW >= c->chrSrcW &&

  1363.                              (srcW & 15) == 0;

  1364.         if (!c->canMMXEXTBeUsed && dstW >= srcW && c->chrDstW >= c->chrSrcW && (srcW & 15) == 0

  1365. 

  1366.             && (flags & SWS_FAST_BILINEAR)) {

  1367.             if (flags & SWS_PRINT_INFO)

  1368.                 av_log(c, AV_LOG_INFO,

  1369.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1370.         }

  1371.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1372.             c->canMMXEXTBeUsed = 0;

  1373.     } else

  1374.         c->canMMXEXTBeUsed = 0;

  1375. 

  1376.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1377.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1378. 

  1379.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1380.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1381.      * correct scaling.

  1382.      * n-2 is the last chrominance sample available.

  1383.      * This is not perfect, but no one should notice the difference, the more

  1384.      * correct variant would be like the vertical one, but that would require

  1385.      * some special code for the first and last pixel */

  1386.     if (flags & SWS_FAST_BILINEAR) {

  1387.         if (c->canMMXEXTBeUsed) {

  1388.             c->lumXInc += 20;

  1389.             c->chrXInc += 20;

  1390.         }

  1391.         // we don't use the x86 asm scaler if MMX is available

  1392.         else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {

  1393.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1394.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1395.         }

  1396.     }

  1397. 

  1398. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1399. 

  1400.     /* precalculate horizontal scaler filter coefficients */

  1401.     {

  1402. #if HAVE_MMXEXT_INLINE

  1403. // can't downscale !!!

  1404.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1405.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1406.                                                              NULL, NULL, 8);

  1407.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1408.                                                              NULL, NULL, NULL, 4);

  1409. 

  1410. #if USE_MMAP

  1411.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1412.                                           PROT_READ | PROT_WRITE,

  1413.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1414.                                           -1, 0);

  1415.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1416.                                           PROT_READ | PROT_WRITE,

  1417.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1418.                                           -1, 0);

  1419. #elif HAVE_VIRTUALALLOC

  1420.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1421.                                                   c->lumMmxextFilterCodeSize,

  1422.                                                   MEM_COMMIT,

  1423.                                                   PAGE_EXECUTE_READWRITE);

  1424.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1425.                                                   c->chrMmxextFilterCodeSize,

  1426.                                                   MEM_COMMIT,

  1427.                                                   PAGE_EXECUTE_READWRITE);

  1428. #else

  1429.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1430.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1431. #endif

  1432. 

  1433. #ifdef MAP_ANONYMOUS

  1434.             if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)

  1435. #else

  1436.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1437. #endif

  1438.             {

  1439.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1440.                 return AVERROR(ENOMEM);

  1441.             }

  1442. 

  1443.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1444.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1445.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1446.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1447. 

  1448.             init_hscaler_mmxext(      dstW, c->lumXInc, c->lumMmxextFilterCode,

  1449.                                 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1450.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1451.                                 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1452. 

  1453. #if USE_MMAP

  1454.             if (   mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1

  1455.                 || mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1) {

  1456.                 av_log(c, AV_LOG_ERROR, "mprotect failed, cannot use fast bilinear scaler\n");

  1457.                 goto fail;

  1458.             }

  1459. #endif

  1460.         } else

  1461. #endif /* HAVE_MMXEXT_INLINE */

  1462.         {

  1463.             const int filterAlign = X86_MMX(cpu_flags)     ? 4 :

  1464.                                     PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1465. 

  1466.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1467.                            &c->hLumFilterSize, c->lumXInc,

  1468.                            srcW, dstW, filterAlign, 1 << 14,

  1469.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1470.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1471.                            c->param,

  1472.                            get_local_pos(c, 0, 0, 0),

  1473.                            get_local_pos(c, 0, 0, 0)) < 0)

  1474.                 goto fail;

  1475.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1476.                            &c->hChrFilterSize, c->chrXInc,

  1477.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1478.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1479.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1480.                            c->param,

  1481.                            get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),

  1482.                            get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0)) < 0)

  1483.                 goto fail;

  1484.         }

  1485.     } // initialize horizontal stuff

  1486. 

  1487.     /* precalculate vertical scaler filter coefficients */

  1488.     {

  1489.         const int filterAlign = X86_MMX(cpu_flags)     ? 2 :

  1490.                                 PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1491. 

  1492.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1493.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1494.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1495.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1496.                        c->param,

  1497.                        get_local_pos(c, 0, 0, 1),

  1498.                        get_local_pos(c, 0, 0, 1)) < 0)

  1499.             goto fail;

  1500.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1501.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1502.                        filterAlign, (1 << 12),

  1503.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1504.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1505.                        c->param,

  1506.                        get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),

  1507.                        get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)

  1508. 

  1509.             goto fail;

  1510. 

  1511. #if HAVE_ALTIVEC

  1512.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1513.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1514. 

  1515.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1516.             int j;

  1517.             short *p = (short *)&c->vYCoeffsBank[i];

  1518.             for (j = 0; j < 8; j++)

  1519.                 p[j] = c->vLumFilter[i];

  1520.         }

  1521. 

  1522.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1523.             int j;

  1524.             short *p = (short *)&c->vCCoeffsBank[i];

  1525.             for (j = 0; j < 8; j++)

  1526.                 p[j] = c->vChrFilter[i];

  1527.         }

  1528. #endif

  1529.     }

  1530. 

  1531.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1532.     c->vLumBufSize = c->vLumFilterSize;

  1533.     c->vChrBufSize = c->vChrFilterSize;

  1534.     for (i = 0; i < dstH; i++) {

  1535.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1536.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1537.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1538.                                << c->chrSrcVSubSample));

  1539. 

  1540.         nextSlice >>= c->chrSrcVSubSample;

  1541.         nextSlice <<= c->chrSrcVSubSample;

  1542.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1543.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1544.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1545.             (nextSlice >> c->chrSrcVSubSample))

  1546.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1547.                              c->vChrFilterPos[chrI];

  1548.     }

  1549. 

  1550.     for (i = 0; i < 4; i++)

  1551.         FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);

  1552. 

  1553.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1554.      * need to allocate several megabytes to handle all possible cases) */

  1555.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1556.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1557.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1558.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1559.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1560.     /* Note we need at least one pixel more at the end because of the MMX code

  1561.      * (just in case someone wants to replace the 4000/8000). */

  1562.     /* align at 16 bytes for AltiVec */

  1563.     for (i = 0; i < c->vLumBufSize; i++) {

  1564.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1565.                           dst_stride + 16, fail);

  1566.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1567.     }

  1568.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1569.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1570.     c->uv_offx2 = dst_stride + 16;

  1571.     for (i = 0; i < c->vChrBufSize; i++) {

  1572.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1573.                          dst_stride * 2 + 32, fail);

  1574.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1575.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1576.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1577.     }

  1578.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1579.         for (i = 0; i < c->vLumBufSize; i++) {

  1580.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1581.                               dst_stride + 16, fail);

  1582.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1583.         }

  1584. 

  1585.     // try to avoid drawing green stuff between the right end and the stride end

  1586.     for (i = 0; i < c->vChrBufSize; i++)

  1587.         if(desc_dst->comp[0].depth_minus1 == 15){

  1588.             av_assert0(c->dstBpc > 14);

  1589.             for(j=0; j<dst_stride/2+1; j++)

  1590.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1591.         } else

  1592.             for(j=0; j<dst_stride+1; j++)

  1593.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1594. 

  1595.     av_assert0(c->chrDstH <= dstH);

  1596. 

  1597.     if (flags & SWS_PRINT_INFO) {

  1598.         const char *scaler = NULL, *cpucaps;

  1599. 

  1600.         for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {

  1601.             if (flags & scale_algorithms[i].flag) {

  1602.                 scaler = scale_algorithms[i].description;

  1603.                 break;

  1604.             }

  1605.         }

  1606.         if (!scaler)

  1607.             scaler =  "ehh flags invalid?!";

  1608.         av_log(c, AV_LOG_INFO, "%s scaler, from %s to %s%s ",

  1609.                scaler,

  1610.                av_get_pix_fmt_name(srcFormat),

  1611. #ifdef DITHER1XBPP

  1612.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1613.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1614.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1615.                                                              "dithered " : "",

  1616. #else

  1617.                "",

  1618. #endif

  1619.                av_get_pix_fmt_name(dstFormat));

  1620. 

  1621.         if (INLINE_MMXEXT(cpu_flags))

  1622.             cpucaps = "MMXEXT";

  1623.         else if (INLINE_AMD3DNOW(cpu_flags))

  1624.             cpucaps = "3DNOW";

  1625.         else if (INLINE_MMX(cpu_flags))

  1626.             cpucaps = "MMX";

  1627.         else if (PPC_ALTIVEC(cpu_flags))

  1628.             cpucaps = "AltiVec";

  1629.         else

  1630.             cpucaps = "C";

  1631. 

  1632.         av_log(c, AV_LOG_INFO, "using %s\n", cpucaps);

  1633. 

  1634.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1635.         av_log(c, AV_LOG_DEBUG,

  1636.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1637.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1638.         av_log(c, AV_LOG_DEBUG,

  1639.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1640.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1641.                c->chrXInc, c->chrYInc);

  1642.     }

  1643. 

  1644.     c->swscale = ff_getSwsFunc(c);

  1645.     return 0;

  1646. fail: // FIXME replace things by appropriate error codes

  1647.     return -1;

  1648. }

  1649. 

  1650. #if FF_API_SWS_GETCONTEXT

  1651. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1652.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1653.                            int flags, SwsFilter *srcFilter,

  1654.                            SwsFilter *dstFilter, const double *param)

  1655. {

  1656.     SwsContext *c;

  1657. 

  1658.     if (!(c = sws_alloc_context()))

  1659.         return NULL;

  1660. 

  1661.     c->flags     = flags;

  1662.     c->srcW      = srcW;

  1663.     c->srcH      = srcH;

  1664.     c->dstW      = dstW;

  1665.     c->dstH      = dstH;

  1666.     c->srcFormat = srcFormat;

  1667.     c->dstFormat = dstFormat;

  1668. 

  1669.     if (param) {

  1670.         c->param[0] = param[0];

  1671.         c->param[1] = param[1];

  1672.     }

  1673. 

  1674.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1675.         sws_freeContext(c);

  1676.         return NULL;

  1677.     }

  1678. 

  1679.     return c;

  1680. }

  1681. #endif

  1682. 

  1683. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1684.                                 float lumaSharpen, float chromaSharpen,

  1685.                                 float chromaHShift, float chromaVShift,

  1686.                                 int verbose)

  1687. {

  1688.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1689.     if (!filter)

  1690.         return NULL;

  1691. 

  1692.     if (lumaGBlur != 0.0) {

  1693.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1694.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1695.     } else {

  1696.         filter->lumH = sws_getIdentityVec();

  1697.         filter->lumV = sws_getIdentityVec();

  1698.     }

  1699. 

  1700.     if (chromaGBlur != 0.0) {

  1701.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1702.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1703.     } else {

  1704.         filter->chrH = sws_getIdentityVec();

  1705.         filter->chrV = sws_getIdentityVec();

  1706.     }

  1707. 

  1708.     if (chromaSharpen != 0.0) {

  1709.         SwsVector *id = sws_getIdentityVec();

  1710.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1711.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1712.         sws_addVec(filter->chrH, id);

  1713.         sws_addVec(filter->chrV, id);

  1714.         sws_freeVec(id);

  1715.     }

  1716. 

  1717.     if (lumaSharpen != 0.0) {

  1718.         SwsVector *id = sws_getIdentityVec();

  1719.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1720.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1721.         sws_addVec(filter->lumH, id);

  1722.         sws_addVec(filter->lumV, id);

  1723.         sws_freeVec(id);

  1724.     }

  1725. 

  1726.     if (chromaHShift != 0.0)

  1727.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1728. 

  1729.     if (chromaVShift != 0.0)

  1730.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1731. 

  1732.     sws_normalizeVec(filter->chrH, 1.0);

  1733.     sws_normalizeVec(filter->chrV, 1.0);

  1734.     sws_normalizeVec(filter->lumH, 1.0);

  1735.     sws_normalizeVec(filter->lumV, 1.0);

  1736. 

  1737.     if (verbose)

  1738.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1739.     if (verbose)

  1740.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1741. 

  1742.     return filter;

  1743. }

  1744. 

  1745. SwsVector *sws_allocVec(int length)

  1746. {

  1747.     SwsVector *vec;

  1748. 

  1749.     if(length <= 0 || length > INT_MAX/ sizeof(double))

  1750.         return NULL;

  1751. 

  1752.     vec = av_malloc(sizeof(SwsVector));

  1753.     if (!vec)

  1754.         return NULL;

  1755.     vec->length = length;

  1756.     vec->coeff  = av_malloc(sizeof(double) * length);

  1757.     if (!vec->coeff)

  1758.         av_freep(&vec);

  1759.     return vec;

  1760. }

  1761. 

  1762. SwsVector *sws_getGaussianVec(double variance, double quality)

  1763. {

  1764.     const int length = (int)(variance * quality + 0.5) | 1;

  1765.     int i;

  1766.     double middle  = (length - 1) * 0.5;

  1767.     SwsVector *vec;

  1768. 

  1769.     if(variance < 0 || quality < 0)

  1770.         return NULL;

  1771. 

  1772.     vec = sws_allocVec(length);

  1773. 

  1774.     if (!vec)

  1775.         return NULL;

  1776. 

  1777.     for (i = 0; i < length; i++) {

  1778.         double dist = i - middle;

  1779.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1780.                         sqrt(2 * variance * M_PI);

  1781.     }

  1782. 

  1783.     sws_normalizeVec(vec, 1.0);

  1784. 

  1785.     return vec;

  1786. }

  1787. 

  1788. SwsVector *sws_getConstVec(double c, int length)

  1789. {

  1790.     int i;

  1791.     SwsVector *vec = sws_allocVec(length);

  1792. 

  1793.     if (!vec)

  1794.         return NULL;

  1795. 

  1796.     for (i = 0; i < length; i++)

  1797.         vec->coeff[i] = c;

  1798. 

  1799.     return vec;

  1800. }

  1801. 

  1802. SwsVector *sws_getIdentityVec(void)

  1803. {

  1804.     return sws_getConstVec(1.0, 1);

  1805. }

  1806. 

  1807. static double sws_dcVec(SwsVector *a)

  1808. {

  1809.     int i;

  1810.     double sum = 0;

  1811. 

  1812.     for (i = 0; i < a->length; i++)

  1813.         sum += a->coeff[i];

  1814. 

  1815.     return sum;

  1816. }

  1817. 

  1818. void sws_scaleVec(SwsVector *a, double scalar)

  1819. {

  1820.     int i;

  1821. 

  1822.     for (i = 0; i < a->length; i++)

  1823.         a->coeff[i] *= scalar;

  1824. }

  1825. 

  1826. void sws_normalizeVec(SwsVector *a, double height)

  1827. {

  1828.     sws_scaleVec(a, height / sws_dcVec(a));

  1829. }

  1830. 

  1831. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1832. {

  1833.     int length = a->length + b->length - 1;

  1834.     int i, j;

  1835.     SwsVector *vec = sws_getConstVec(0.0, length);

  1836. 

  1837.     if (!vec)

  1838.         return NULL;

  1839. 

  1840.     for (i = 0; i < a->length; i++) {

  1841.         for (j = 0; j < b->length; j++) {

  1842.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1843.         }

  1844.     }

  1845. 

  1846.     return vec;

  1847. }

  1848. 

  1849. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1850. {

  1851.     int length = FFMAX(a->length, b->length);

  1852.     int i;

  1853.     SwsVector *vec = sws_getConstVec(0.0, length);

  1854. 

  1855.     if (!vec)

  1856.         return NULL;

  1857. 

  1858.     for (i = 0; i < a->length; i++)

  1859.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1860.     for (i = 0; i < b->length; i++)

  1861.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1862. 

  1863.     return vec;

  1864. }

  1865. 

  1866. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1867. {

  1868.     int length = FFMAX(a->length, b->length);

  1869.     int i;

  1870.     SwsVector *vec = sws_getConstVec(0.0, length);

  1871. 

  1872.     if (!vec)

  1873.         return NULL;

  1874. 

  1875.     for (i = 0; i < a->length; i++)

  1876.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1877.     for (i = 0; i < b->length; i++)

  1878.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1879. 

  1880.     return vec;

  1881. }

  1882. 

  1883. /* shift left / or right if "shift" is negative */

  1884. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1885. {

  1886.     int length = a->length + FFABS(shift) * 2;

  1887.     int i;

  1888.     SwsVector *vec = sws_getConstVec(0.0, length);

  1889. 

  1890.     if (!vec)

  1891.         return NULL;

  1892. 

  1893.     for (i = 0; i < a->length; i++) {

  1894.         vec->coeff[i + (length    - 1) / 2 -

  1895.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1896.     }

  1897. 

  1898.     return vec;

  1899. }

  1900. 

  1901. void sws_shiftVec(SwsVector *a, int shift)

  1902. {

  1903.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1904.     av_free(a->coeff);

  1905.     a->coeff  = shifted->coeff;

  1906.     a->length = shifted->length;

  1907.     av_free(shifted);

  1908. }

  1909. 

  1910. void sws_addVec(SwsVector *a, SwsVector *b)

  1911. {

  1912.     SwsVector *sum = sws_sumVec(a, b);

  1913.     av_free(a->coeff);

  1914.     a->coeff  = sum->coeff;

  1915.     a->length = sum->length;

  1916.     av_free(sum);

  1917. }

  1918. 

  1919. void sws_subVec(SwsVector *a, SwsVector *b)

  1920. {

  1921.     SwsVector *diff = sws_diffVec(a, b);

  1922.     av_free(a->coeff);

  1923.     a->coeff  = diff->coeff;

  1924.     a->length = diff->length;

  1925.     av_free(diff);

  1926. }

  1927. 

  1928. void sws_convVec(SwsVector *a, SwsVector *b)

  1929. {

  1930.     SwsVector *conv = sws_getConvVec(a, b);

  1931.     av_free(a->coeff);

  1932.     a->coeff  = conv->coeff;

  1933.     a->length = conv->length;

  1934.     av_free(conv);

  1935. }

  1936. 

  1937. SwsVector *sws_cloneVec(SwsVector *a)

  1938. {

  1939.     SwsVector *vec = sws_allocVec(a->length);

  1940. 

  1941.     if (!vec)

  1942.         return NULL;

  1943. 

  1944.     memcpy(vec->coeff, a->coeff, a->length * sizeof(*a->coeff));

  1945. 

  1946.     return vec;

  1947. }

  1948. 

  1949. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1950. {

  1951.     int i;

  1952.     double max = 0;

  1953.     double min = 0;

  1954.     double range;

  1955. 

  1956.     for (i = 0; i < a->length; i++)

  1957.         if (a->coeff[i] > max)

  1958.             max = a->coeff[i];

  1959. 

  1960.     for (i = 0; i < a->length; i++)

  1961.         if (a->coeff[i] < min)

  1962.             min = a->coeff[i];

  1963. 

  1964.     range = max - min;

  1965. 

  1966.     for (i = 0; i < a->length; i++) {

  1967.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1968.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1969.         for (; x > 0; x--)

  1970.             av_log(log_ctx, log_level, " ");

  1971.         av_log(log_ctx, log_level, "|\n");

  1972.     }

  1973. }

  1974. 

  1975. void sws_freeVec(SwsVector *a)

  1976. {

  1977.     if (!a)

  1978.         return;

  1979.     av_freep(&a->coeff);

  1980.     a->length = 0;

  1981.     av_free(a);

  1982. }

  1983. 

  1984. void sws_freeFilter(SwsFilter *filter)

  1985. {

  1986.     if (!filter)

  1987.         return;

  1988. 

  1989.     sws_freeVec(filter->lumH);

  1990.     sws_freeVec(filter->lumV);

  1991.     sws_freeVec(filter->chrH);

  1992.     sws_freeVec(filter->chrV);

  1993.     av_free(filter);

  1994. }

  1995. 

  1996. void sws_freeContext(SwsContext *c)

  1997. {

  1998.     int i;

  1999.     if (!c)

  2000.         return;

  2001. 

  2002.     if (c->lumPixBuf) {

  2003.         for (i = 0; i < c->vLumBufSize; i++)

  2004.             av_freep(&c->lumPixBuf[i]);

  2005.         av_freep(&c->lumPixBuf);

  2006.     }

  2007. 

  2008.     if (c->chrUPixBuf) {

  2009.         for (i = 0; i < c->vChrBufSize; i++)

  2010.             av_freep(&c->chrUPixBuf[i]);

  2011.         av_freep(&c->chrUPixBuf);

  2012.         av_freep(&c->chrVPixBuf);

  2013.     }

  2014. 

  2015.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  2016.         for (i = 0; i < c->vLumBufSize; i++)

  2017.             av_freep(&c->alpPixBuf[i]);

  2018.         av_freep(&c->alpPixBuf);

  2019.     }

  2020. 

  2021.     for (i = 0; i < 4; i++)

  2022.         av_freep(&c->dither_error[i]);

  2023. 

  2024.     av_freep(&c->vLumFilter);

  2025.     av_freep(&c->vChrFilter);

  2026.     av_freep(&c->hLumFilter);

  2027.     av_freep(&c->hChrFilter);

  2028. #if HAVE_ALTIVEC

  2029.     av_freep(&c->vYCoeffsBank);

  2030.     av_freep(&c->vCCoeffsBank);

  2031. #endif

  2032. 

  2033.     av_freep(&c->vLumFilterPos);

  2034.     av_freep(&c->vChrFilterPos);

  2035.     av_freep(&c->hLumFilterPos);

  2036.     av_freep(&c->hChrFilterPos);

  2037. 

  2038. #if HAVE_MMX_INLINE

  2039. #if USE_MMAP

  2040.     if (c->lumMmxextFilterCode)

  2041.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  2042.     if (c->chrMmxextFilterCode)

  2043.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  2044. #elif HAVE_VIRTUALALLOC

  2045.     if (c->lumMmxextFilterCode)

  2046.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  2047.     if (c->chrMmxextFilterCode)

  2048.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  2049. #else

  2050.     av_free(c->lumMmxextFilterCode);

  2051.     av_free(c->chrMmxextFilterCode);

  2052. #endif

  2053.     c->lumMmxextFilterCode = NULL;

  2054.     c->chrMmxextFilterCode = NULL;

  2055. #endif /* HAVE_MMX_INLINE */

  2056. 

  2057.     av_freep(&c->yuvTable);

  2058.     av_freep(&c->formatConvBuffer);

  2059. 

  2060.     av_free(c);

  2061. }

  2062. 

  2063. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  2064.                                         int srcH, enum AVPixelFormat srcFormat,

  2065.                                         int dstW, int dstH,

  2066.                                         enum AVPixelFormat dstFormat, int flags,

  2067.                                         SwsFilter *srcFilter,

  2068.                                         SwsFilter *dstFilter,

  2069.                                         const double *param)

  2070. {

  2071.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  2072.                                              SWS_PARAM_DEFAULT };

  2073. 

  2074.     if (!param)

  2075.         param = default_param;

  2076. 

  2077.     if (context &&

  2078.         (context->srcW      != srcW      ||

  2079.          context->srcH      != srcH      ||

  2080.          context->srcFormat != srcFormat ||

  2081.          context->dstW      != dstW      ||

  2082.          context->dstH      != dstH      ||

  2083.          context->dstFormat != dstFormat ||

  2084.          context->flags     != flags     ||

  2085.          context->param[0]  != param[0]  ||

  2086.          context->param[1]  != param[1])) {

  2087.         sws_freeContext(context);

  2088.         context = NULL;

  2089.     }

  2090. 

  2091.     if (!context) {

  2092.         if (!(context = sws_alloc_context()))

  2093.             return NULL;

  2094.         context->srcW      = srcW;

  2095.         context->srcH      = srcH;

  2096.         context->srcFormat = srcFormat;

  2097.         context->dstW      = dstW;

  2098.         context->dstH      = dstH;

  2099.         context->dstFormat = dstFormat;

  2100.         context->flags     = flags;

  2101.         context->param[0]  = param[0];

  2102.         context->param[1]  = param[1];

  2103.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  2104.             sws_freeContext(context);

  2105.             return NULL;

  2106.         }

  2107.     }

  2108.     return context;

  2109. }